With the development of artificial intelligence, many people are no strangers to voiceprint recognition. Voiceprint recognition is to convert sound signals into electrical signals, and then use a computer for identification. Different tasks and applications will use different voiceprint recognition technologies. For example, identification technology may be required when narrowing the scope of criminal investigations, while verification technology may be required for banking transactions.

The noise-cancelling headset 4 Pro recently launched by Xiaomi adopts the bone voiceprint noise reduction method and supports the call noise reduction function. In fact, the bone voiceprint has been released on Huawei FreeBuds 3 as early as 2020. But many people still don't know what is bone voiceprint recognition. And what advantages does bone voiceprint recognition have over our ordinary voiceprint recognition? How it is applied to headphone?

Bone voiceprint recognition

Due to the different conduction mediums, the voice characteristics recorded in the sound conduction are also different. Human bones can conduct sound. Generally, the timbre of our own voice is different from that of others. It is because the voice we hear by ourselves enters our ears through air conduction and bone conduction, while others hear it. We hear our voices through the air. Bone conduction is not as easy as air transmission, and the resistance it produces will cause the frequency of the sound wave to drop, so the tone will become deeper. The diagram below depicts the conduction from bone conduction earphones to the cochlea:

What is bone voiceprint recognition, or to be more specifically, what is voiceprint recognition based on bone conduction speech? Just like human DNA and fingerprints, each person's skull is unique, so the reflection between the bones of the sound is also unique, and each person's bone conduction speech reflection is different. Once collected with the headset, it can be used to identify the person. At present, the headsets that support bone conduction voiceprint mainly include Nanka Runner pro3, Shaoyin OpenRun Pro sports headset, Bone listening SS900 TWS, Aike IKKO ITG01 bone conduction headset, Sony Ericsson GD06, Lenovo X4 and the recent Huawei FreeBuds 3 and Xiaomi drop. Noise Headphones 4 Pro. These bone conduction earphones have been monopolized by foreign bone conduction sensors before. At present, Huawei uses Goertek bone conduction sensors to replace foreign devices, which indicates that domestic bone conduction earphones may occupy more market share in the future.

Advantages of bone voiceprint recognition

Compared with the ordinary voiceprint recognition under air, the advantages of bone conduction voiceprint recognition are as follows:

(1) Since each person's skull is unique, and each person's timbre is also unique, the double judgment of the two has higher accuracy.

(2) Bone conduction voiceprint recognition is less disturbed by reverberation noise, reducing the impact of noise on the recognition accuracy, and reducing the problem of domain mismatch in voiceprint recognition, so it can bring higher recognition accuracy.

Disadvantages of bone voiceprint recognition

For ordinary voiceprint recognition, bone conduction voiceprint recognition needs to wear wireless headphones or microphones, the foreign body sensation is obvious, the price is more expensive than ordinary headphones, and not everyone likes to wear headphones. It needs to be worn every time for voiceprint authentication, which is not convenient enough and has limited application scenarios. The instructions between users and smart home devices are usually random and Under-Air communication methods. Letting users wear headphones to communicate with them may not be in line with user habits and application scenarios, so the market is limited. In addition, there are currently no open source bone conduction and voice data sets on the market, and data needs to be recorded for product applications. Compared with ordinary voiceprint recognition recording, the cost of bone conduction voice recording is higher, and a more professional team is required for support.

About Magic Data

Magic Data is a global AI data solutions provider with 3 core products: MD training datasets, Annotator platform, and data collection and annotation services.

Founded in 2016, Magic Data is dedicated to provide high quality text, audio, image, and multi-modal training datasets to the AI industry. The company now has accumulation of over 400 licensable datasets that are ready-to-use for machine learning, including over 200,000 hours speech data for ASR model and over 3,000,000 sentences of text data for NLP.

Magic Data is ISO/IEC 27001 & ISO/IEC 27701:2019 accredited and GDPR compliant.

For more information, visit www.magicdatatech.com.